0

我正在尝试使用 keras 功能 API 构建神经网络并训练我使用 keras 调谐器的网络。该模型由一些嵌入层和一些密集层组成：

import pandas as pd
import numpy as np
from keras.models import Sequential
from keras.layers import Input, Embedding, Dense, Flatten
from tensorflow import keras
from sklearn.preprocessing import LabelEncoder
from tensorflow.keras.utils import plot_model
import matplotlib.pyplot as plt
from kerastuner.tuners import RandomSearch, BayesianOptimization

def build_model(hp):
  model = keras.Sequential()
  activation = hp.Choice('activation',['relu','tanh','linear'])

  optimizer = hp.Choice('optimizer', ['adam', 'sgd', 'rmsprop'])
  in_layers = list()
  em_layers = list()
  for i in range(len(X_train_enc)):
    # calculate the number of unique inputs
    n_labels = len(np.unique(X_train_enc[i]))
    # define input layer
    in_layer = Input(shape=(1,))
    # define embedding layer
    em_layer = Embedding(n_labels, round(n_labels/2))(in_layer)
    # store layers
    in_layers.append(in_layer)
    em_layers.append(em_layer)

  merge = keras.layers.concatenate(em_layers)
  x = Flatten()(merge)
  
  for i in range(hp.Int('num_layers', 1, 6)):
    units = hp.Int(
          'units_' + str(i),
          min_value=8,
          max_value=128,
          step=16
    )
    x = Dense(units, activation=activation)(x)
    drop_rate = hp.Choice('drop_rate_' + str(i),
                            [
                              0.0, 0.1, 0.2, 0.3, 0.4,
                              0.5, 0.6, 0.7, 0.8, 0.9
                            ])
    x = keras.layers.Dropout(rate=drop_rate)(x)


  output = Dense(1, activation='linear')(x)
  model = keras.models.Model(inputs=in_layers, outputs=output)

  model.compile(
      optimizer=optimizer,
      loss=keras.losses.MeanSquaredError(reduction="auto", name="mean_squared_error"), 
      metrics=['accuracy']
  )
  return model

要使用调谐器，请执行以下代码：

tuner = BayesianOptimization(
    build_model,
    objective='accuracy',
    max_trials=25,
    executions_per_trial=5,
    directory='drive/MyDrive/Master/train_model/nn_first_reg',
    project_name='nn_bayes_first_reg',
    seed=10)


tuner.search(X_train_enc,y_train)
tuner.results_summary()
best_hyperparameters = tuner.get_best_hyperparameters(1)[0]
model = tuner.hypermodel.build(best_hyperparameters)
history = model.fit(X_train_enc, y_train, epochs=50, validation_split=0.2)
val_acc_per_epoch = history.history['val_accuracy']
best_epoch = val_acc_per_epoch.index(max(val_acc_per_epoch)) + 1
print('Best epoch: %d' % (best_epoch,))

调谐器的结果用于构建模型。

_, accuracy = model.evaluate(X_test_enc, y_test)
print('Accuracy: %.2f' % (accuracy*100))
print(model.summary())
from tensorflow.keras.utils import plot_model
plot_model(model, to_file='drive/MyDrive/Master/plots/results/before_game/nn_first.png')

model.summary()和的输出plot_model为空。输出如下所示：

Model: "model"
__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
Total params: 137,329
Trainable params: 137,329
Non-trainable params: 0
__________________________________________________________________________________________________
None

不知道这是否已连接，但代码运行得非常好，我只是绘制网络以查看它是如何构建的。但是，当我尝试保存模型时，我确实遇到了错误：

model.save('drive/MyDrive/Master/SavedModels/nn_first_before.csv')

错误是：

KeyError: 'input_269_ib-0'

我不知道这些是否相关。

4

1 回答 1

1

tf.keras您正在混合来自和库的导入keras，它们不是同一个库并且不兼容，从而产生了您在此处看到的奇怪问题。注意你的进口：

from keras.models import Sequential
from keras.layers import Input, Embedding, Dense, Flatten
from tensorflow import keras
from sklearn.preprocessing import LabelEncoder
from tensorflow.keras.utils import plot_model

在这里，您使用库中的一些图层，以及keras库中的其他图层tensorflow.keras，这将不起作用。仅使用来自这些库之一的导入。

于 2021-06-06T22:28:09.683 回答